This interesting paper uses causal mediation analysis and a sample of Australian children in order to examine whether the relationship between disability and mental health is mediated by bullying. In addition, the authors sought to investigate whether dependent measurement error impacted their findings. The primary results were that a substantial portion of the disability-mental health relationship is mediated by bullying, a finding which appears to be robust to dependent measurement error. Notable strengths include the longitudinal study design and use of the counterfactual and potential outcomes approaches to mediation analysis.

Overall, the paper is well organized and coherent. The introduction provides sufficient rationale, clearly outlines what is known about the topic, and identifies the knowledge gaps which this paper seeks to address. A unique feature of this study is the use of multi-informants in order to evaluate the potential for dependent measurement error. Examining whether findings hold with the use of different reporters (e.g., adolescent or parent), is comprehensive and consistent with clinical practice in psychology, which increases the study’s clinical relevance. The included figures are a nice visual aid for the reader, and the paper concludes with important limitations of the studied exposure and mediator. That said, there are some important weaknesses which need to be addressed: the temporal overlap between the mediator and the outcome, the lack of information on those lost to follow-up, and the potential for increased reporting of bullying among adolescents with disabilities and their parents/caregivers.

More specific feedback for the authors is provided below.

Major comments

-Given the nature of the causal inferences being drawn from this study, I am concerned about appropriate temporal ordering of the mediator and outcome. Since both were measured at wave 6, it is possible that poor mental health influences the perception of bullying, as opposed to how the authors have conceptualized the model.

-Related, the authors excluded 31.6% of the original sample, as these were the persons not surveyed at waves 4-6. It would be helpful for the reader to be able to understand whether those who participated in the studied survey cycles were similar, socio-demographically speaking, to those lost to follow-up.

-It is possible that recall bias could have been introduced as a result of the exposure under study (disability). Parents of children with a disability may be more likely to be sensitive to and report bullying related events. Given that parental report of bullying is the preferred mediator under study, recall bias may have impacted the primary study findings.

Minor comments

-In the abstract, it would have been helpful for the reader to have information on the specifics of the analytic method used (e.g., parametric regression based approach) to test the causal mediation analysis.

-The authors noted that the SDQ (measure of mental health) is validated in adolescents, even those with an intellectual disability. This is an important consideration given the nature of the exposure, however, in the supplemental appendix A, I see that one of the conditions listed was “Difficulty learning or understanding things”. Such difficulties could be the result of an intellectual disability, as the authors were rightly concerned about, or other diagnoses such as attention deficit hyperactivity disorder. Given that the SDQ taps into aspects of attention deficit hyperactivity disorder (i.e., inattention and hyperactivity) and peer related issues (as per the author’s reference #29), can the authors be sure that there isn’t some degree of overlap between the outcome, exposure and/or mediator?

-In terms of important confounders not included, did the authors have access to information on child obesity, stigma, or social support? These are potential confounders which may impact the associations of interest.

-Parental K6 scores were included as a measure of depression. To my knowledge, the K6 measures non-specific psychological distress, including both depressive and anxious symptoms?

-Why did the authors constrain parental education to only two levels, and not include a level which captures post-secondary achievement?

-The total amount of missing data was relatively large (16.8%), even though amounts of missingness by each individual variable was within acceptable limits (<5% for all but one measure). The authors should consider using multiple imputation as part of a sensitivity analysis in order to assess the impact of missing data on their primary model of interest.

-In the results section, it is unclear how the decision to retain which covariates in the analyses was made. Was this decision based on statistical significance? For example, it appears that there are differences in parental depression across groups (Tables 1 and 2), but this variable was not mentioned as being included in the analyses (see results section, paragraph #2).

-While the reader can compute the proportion mediated from the information presented in Table 3, the results would be clearer if there was information on how the proportion was computed in the methods section. This would be especially helpful to those readers unfamiliar with mediational analyses.

-The first line of the last paragraph (in the results section) should explain that Figure 4 shows the "results" of the analyses, not the “analyses” themselves. Also in this paragraph, the authors note that the effect sizes from the total effect estimates were compared to the “…SD of the overall sample mean for the preferred model…”. This is important information that helps with interpreting the findings, and could appear in the methods section, or explained earlier in the results.

-The authors note that their work advances the work of others as they use a modeling method which can account for non-linearity and interaction between the exposure and the mediator. Was there evidence of either in the present study?

-Consideration should be given as to whether the inclusion of other factors on the causal pathway from bullying to mental health (e.g., stigma, coping abilities) could be conceptualized in a multiple mediator model. If so, would their inclusion reduce the amount of the relationship mediated specifically by bullying? In which case, some may be worthy of mention as they may be amenable to change.